Methods, systems, and computer products for evaluating robustness of a list scheduling framework

Abstract

Systems, methods, and computer products for evaluating robustness of a list scheduling framework. Exemplary embodiments include a method for evaluating the robustness of a list scheduling framework, the method including identifying a set of compiler benchmarks known to be sensitive to an instruction scheduler, running the set of benchmarks against a heuristic under test, H and collect an execution time Exec(H[G]), where G is a directed a-cyclical graph, running the set of benchmarks against a plurality of random heuristics Hrand[G]i, and collect a plurality of respective execution times Exec(Hrand[G])i, computing a robustness of the list scheduling framework, and checking robustness check it against a pre-determined threshold.

Description

TRADEMARKS[0001]IBM® is a registered trademark of International Business Machines Corporation, Armonk, N.Y., U.S.A. Other names used herein may be registered trademarks, trademarks or product names of International Business Machines Corporation or other companies.BACKGROUND OF THE INVENTION[0002]1. Field of the Invention[0003]This invention relates to compilers, and particularly to systems, methods, and computer products for evaluating robustness of a list scheduling framework.[0004]2. Description of Background[0005]Performance problems in compilers are generally difficult to identify and diagnose. Newly introduced performance regressions could go unnoticed for extended periods of time unless great care is taken with performance regression testing. In practice, however, unmanned automated performance regression testing of compiler components continues to be a difficult problem. Furthermore, with increasing complexity of underlying hardware pipelines and memory hierarchies, failing t...

AI Technical Summary

Benefits of technology

[0011]As a result of the summarized invention, technically we have achieved a solution which a framework is automated for performa...

Problems solved by technology

Performance problems in compilers are generally difficult to identify and diagnose.

In practice, however, unmanned automated performance regression testing of compiler components continues to be a difficult problem.

Furthermore, with increasing complexity of underlying hardware pipelines and memory hierarchies, failing to model these complex structures can result in generated code being significantly faster or slower than expected.

Images